OpenAI has introduced Lockdown Mode, a new feature aimed at enhancing protection against prompt injection attacks, which involve embedding malicious instructions in web content. The mode disables live web browsing and image retrieval from the internet, while allowing users to generate images. It also restricts deep research and agent mode functionalities. According to OpenAI, even with Lockdown Mode activated, ChatGPT may still be vulnerable to prompt injections that could influence response accuracy or behavior. However, the feature is intended to minimize the risk of sensitive data exposure during such attacks.
Lockdown Mode is specifically targeted at users and organizations handling sensitive data, as stated by OpenAI. The company emphasized that the mode is not suitable for all users and is being rolled out to eligible ChatGPT Business accounts and personal accounts. The feature aims to reduce the likelihood of data exfiltration through prompt injection attacks by limiting access to external content sources. OpenAI noted that prompt injections could still appear in cached web content or uploaded files, potentially affecting response behavior or accuracy.
OpenAI announced the release of Lockdown Mode on June 6, 2026, as part of its ongoing efforts to improve security measures for its AI models. The company highlighted the importance of the feature in safeguarding sensitive data from potential threats posed by prompt injection attacks. The rollout is currently limited to specific user groups, reflecting the targeted nature of the security enhancement.
Source: techcrunch
